Philosophy

Philosophy

isotop has no practical use. A process list that did the job would be top, which exists, is faster, and fits in a corner of the screen. isotop is instead a machine for turning the hum of an operating system into pictures that are worth looking at for their own sake, and for taking the physics, mathematics and biology behind those pictures seriously enough to write them down. This wiki is where that writing lives.

It is maximalist on purpose. There are twenty-one views and eight more specified, a CPU rasterizer and a GPU rasterizer that must agree pixel for pixel, a Fourier transform written in-tree so a planet can carry an epicycle chain, a Lorentz boost applied to the history of a shell. None of that is proportionate to the problem of "which process is using the CPU". It is proportionate to a different problem: how much of the structure of a running computer can be made visible, and how beautiful can the visible thing be without lying?

History

The question of what a machine is doing has been asked, with software, for as long as there have been machines to ask it of. The tools that answer it each reflect their decade.

• /proc. Tom Killian built the first process filesystem for Eighth Edition Unix at Bell Labs and described it in "Processes as Files" at the USENIX conference in the summer of 1984. Its purpose was practical: to replace the ptrace system call with files that could be read and written. Linux's /proc, which isotop reads, descends from the idea, not the code.
• top. The first version was written by William LeFebvre and posted to the net.sources newsgroup in 1984. Where it began is disputed: later copyright notices in the code reach back to 1980 and credit an earlier author at Lucasfilm. Either way, a screen of the busiest processes, refreshed in place, has been the default answer for four decades.
• Treemaps. Ben Shneiderman and colleagues at the University of Maryland devised the treemap in the early 1990s (the first papers are dated 1991) to answer a question about disk space: where on this disk did the space go? It is the first widely used picture in which area is a measurement of a resource, and the ancestor of every disk-usage visualizer since.
• Task Manager. The Windows NT Task Manager, which shipped with NT 4.0 in 1996, was written by Dave Plummer at home, as a shareware idea he meant to sell. Dave Cutler, the architect of NT, saw it and agreed to take it into the product, and Plummer donated it.
• htop. Hisham Muhammad released htop in May 2004, as an interactive alternative to top that shows every process, with colour and a mouse.

Each of these was built to be used. A picture of a process table is a different kind of artifact, and the work isotop belongs to is the one that treats the data as something worth looking at: software cities and treemaps, as visualization research, and a long tradition of screensavers and demoscene pieces, as art. isotop is in the second tradition with the discipline of the first. The next section says what the discipline is.

The one rule: the realness contract

The art is allowed to be extravagant. It is not allowed to be fake. Every visual element in a view maps to a measured quantity, or the documentation declares it decorative. Each view's page opens its mapping with a table: visual element, measured source, transform (log, square root, smoothing constant, quantum). When a source cannot be read (permissions, kernel version, hardware), the view degrades visibly and says so in its status strip. Live mode never fills a gap with invented numbers. Simulations advance on wall-clock time, not frames, and their randomness is seeded from process identity so the same machine state draws the same picture.

This is the constraint that makes the project interesting rather than merely pretty. A picture of a bubble chamber is just a picture. A picture of a bubble chamber in which every track's curvature is a measured quantity, and in which you can say exactly which, is a small instrument. The question each page asks is the one a physicist asks of any plot: what does this let me know, and what does it not?

What you can and cannot know

Every view is a lossy channel. A process table sampled once a second cannot see anything faster than a half second; a kernel counter is cumulative, so a rate is a difference of two readings and the first reading has no rate; a value that was not measured is unknown, not zero. Each page therefore has a section on reading the picture: which things are measured, which are derived, which are decorative, and what is invisible. The pages on sampling, the model and determinism are the formal version of the same stance.

Why physics, why biology, why history

The metaphors are not arbitrary. A process is born by fork, lives by competing for a finite resource, and dies leaving a record; so does a particle in a bubble chamber (Planned Chamber), a hypha in a slime mold (Planned Mold), a hen in a coop (View Coop). Memory pressure that ends in the kernel killing something is, at the level of structure, a horizon (View Horizon). The history of astronomy gave us a way of describing any periodic motion as a chain of circles (View Ptolemy) long before Fourier made the idea exact. Where the correspondence is deep, the page says why. Where it is only a pleasing resemblance, the page says that too.

How the wiki is organised

• Theory pages describe how isotop works, from the kernel counters to the pixels in the terminal. Start with the Theory Overview.
• Math pages treat a technique on its own terms, independent of any one view: the wave equation, Fourier epicycles, flocking, reaction-diffusion, and so on.
• View pages cover each implemented view; Planned pages cover the views that are specified but not yet built, theory first, clearly marked as designs.
• Guides hold the practical material: writing a view (there is no dynamic plugin system, and the page says so), recipes, and how to render reproducibly.
• The Gallery is the media; the Glossary is the vocabulary.

Non-goals

isotop does not try to be fast at the expense of being legible, minimal at the expense of being interesting, or useful at the expense of being true. It never sends anything off the machine. It never escalates privilege. It does not pretend a decorative element is a measurement. Beyond that, the design goal is to see how far a terminal, a handful of kernel interfaces and some mathematics can be pushed, and to keep an honest account of what was done and why.